Construction of Three-Dimensional Feature Point Model for Virtual Assembly System using Visual ID Tags

Abstract

This paper proposes a method of developing an object shape model for a virtual assembly system using a combination of visual ID tags and three-dimensional (3D) natural feature points. The object shape model with its real size information is useful in large structures where maintenance robots are used to perform repetitive tasks. We developed the feature-point-based shape model by capturing visual ID tags and the feature points of the image with a monocular camera. The developed model can be used for the detection of the object against a background image and for an estimation of its 3D pose (position and orientation). To estimate the pose of the object using the proposed method, we assigned 3D feature points to the captured image using its scale-invariant feature transform features. The method can be applied to complex background images by using visual ID tags or constructing a feature point model. Our experimental results confirmed the feasibility of the proposed method.